flash-attention
dnabert2-no-flashattentionqwopus3.8-27b-flash-abliterated-attention8-bf16recurrence-vision-mtplxqwopus3.8-27b-flash-attention8-bf16recurrence-vision-mtplxflash-attention-3TinyLlama2-jp-122M-FlashAttention2-GGUFdnaberts-no-flashattentionLlama-2-7b-hf-dialogsum-test-flash-attention-2bark-flash-attention-2
A10_benchmark_flash_attentionFlash-Attention3-wheel
Flash-Attention3-wheel
Flash Attention 3 wheels on commit 0e60e39473e8df549a20fb5353760f7a65b30e2d.
https://windreamer.github.io/flash-attention3-wheels/ much more banger!
Build using H100
For PyTorch 2.6.0 12.6, 2.7.0 12.6, 2.7.0 12.8, 2.7.1 12.6, 2.7.1 12.8, minimum Python 3.9.
Build using GH200 ARM64
For PyTorch 2.7.0 12.8, 2.7.1 12.8, minimum Python 3.9.
Installation
# make sure rename or else your PIP is screaming
# eg remove `-2.7.1-12.8`… See the full description on the dataset page: https://huggingface.co/datasets/malaysia-ai/Flash-Attention3-wheel.flash-attentionflash-attention-windows
Overview
flash-attention の 非公式 Windows 用 wheel です。
StabilityMatrix の ComfyUI 環境で動作するwheelが見当たらなかったため、
公式ソースコードからビルドしました。
Compatible Environment
This wheel is built and tested with the following environment:
Python: 3.10.11
PyTorch: 2.14.0
CUDA: 13.0
License
This repository distributes a compiled wheel derived from flash-attention,
which is licensed under the BSD 3-Clause License.
This project is not affiliated with or endorsed by the original… See the full description on the dataset page: https://huggingface.co/datasets/yshero/flash-attention-windows.flash-attentionflash_attention_2_source